SearchByAI lists endpoints you already run. If you have nothing serving yet, this is the stack we install for customers.
This is a Docker install on a Linux machine you control. Run this first — it reports what is present, what is missing, and which build fits. It installs nothing unless you say yes.
bash <(curl -fsSL https://searchbyai.com/stack/preflight.sh)
Everything splits on one question: do you have a domain?
| A — I have a domain | B — I do not | |
|---|---|---|
| Hostname | Cloudflare tunnel on your domain | Tailscale Funnel, free |
| Cost | ~$10/year | nothing |
| Setup | 10 min, plus up to 24h for DNS | 5 min |
| n8n Google nodes | yes | no — Google will not redirect to .ts.net |
| Everything else | identical | identical |
Path B is a real option, not a downgrade. You lose Gmail, Drive, Sheets and Calendar inside n8n. n8n itself, Open WebUI, LightRAG, crawl4ai, local models and being listed all work the same — and adding a domain later changes nothing else.
One GPU, one command, then a guided credential walkthrough.
| Ollama | Local models, OpenAI-compatible API. Ships with Gemma 4 for chat and nomic-embed-text for retrieval; swap either for anything in the Ollama library. |
|---|---|
| Open WebUI | Chat interface, API keys for your own tools |
| n8n | Workflow automation — this is where endpoints come from |
| LightRAG | Retrieval over your own documents |
| crawl4ai | Web scraping for agents |
| Cloudflare Tunnel | Public HTTPS with no open ports and no static IP |
A GPU is optional — there are two builds.
| gpu | cloud | |
|---|---|---|
| Needs | NVIDIA GPU, 8GB+ VRAM, container toolkit | No GPU |
| Inference | Local models via Ollama | Anthropic, OpenAI or Gemini |
| Downloads | ~9GB of models | Nothing |
There is no CPU-inference option on purpose: a 9GB model answering at a few tokens per second is worse than an API key, and it pins the machine while it runs. Without a GPU you get the cloud build, which ships without Ollama entirely — n8n, Open WebUI, LightRAG and crawl4ai are identical either way. Docker and Linux needed for both.
Credentials, in the one order that works. The installer walks each step with the exact URLs and click paths.
Every credential is generated by you, in your own dashboard, and saved to your own machine. None of it is transmitted to us.
Two ways in. Both end with the same stack running on your own hardware.
Point Claude, ChatGPT or any agent with shell access at the install guide and let it do the work. It runs the commands, and stops to walk you through the five credentials that can only be issued by a human in a browser.
Read https://github.com/aiwtfgpt/searchbyai-public/blob/main/stack/INSTALL.md and install this stack on my machine.
mkdir -p ~/ai-stack && cd ~/ai-stack
curl -fsSL https://raw.githubusercontent.com/aiwtfgpt/searchbyai-public/main/stack/docker-compose.yml -o docker-compose.yml
curl -fsSL https://raw.githubusercontent.com/aiwtfgpt/searchbyai-public/main/stack/scripts/install-stack.sh -o install-stack.sh
bash install-stack.sh
Then follow INSTALL.md from step 2 for the tunnel, Google, n8n and Claude credentials.
Read the install guide Back to the registry
Already have your own stack? You do not need any of this — just list your node.
Tailscale Funnel gives you a free, stable, public HTTPS URL — no domain, no open ports, no router configuration, works behind CGNAT.
curl -fsSL https://tailscale.com/install.sh | sh
sudo tailscale up
sudo tailscale funnel 5678
You get something like
https://your-machine.your-tailnet.ts.net, stable across
restarts. Register that with SearchByAI and you are listed.
The trade: Google's OAuth only redirects to a domain you control, so with Funnel you lose n8n's Gmail, Drive, Sheets, Docs and Calendar nodes. Everything else is identical. Start with Funnel and add a domain later if you want them — nothing else changes.
A Replit app, a Hugging Face Space, a Vercel function — the registry only needs a URL that answers. List it and skip this stack entirely.
Install without the tunnel. Everything runs on
localhost. You cannot be listed, but nothing stops you
building.
A stack nobody can find is just a server. Listing it on SearchByAI is what makes your endpoints discoverable — and it is free.
bash <(curl -fsSL https://searchbyai.com/connect.sh)
Your node shows up in search with what it does, which model runs it, what each endpoint asks for, and uptime measured by our probes rather than self-reported. Agents and people find it and connect straight to your machine — we are the index, never a proxy, so your requests never touch our servers.
After that, sign in with the same email — no password. That is where you edit the listing, add endpoints, and set what you charge. You keep all of it; payment is between you and your customer, and we do not process it.